You are on page 1of 3

.

Roll No. ...................... Total No. of Questions: 08]


cJ!'.
...

[Total No. of Pages: <p' , ' M.Tech.

NEURAL NETWORKS

AND FUZZY LOGICS

SUBJECT CODE: EC - 505 Paper ID: (E0497]


[Note: Please fill subject code and paper ID on OMR]

Time: 03 Hours
Instruction to Candidates: 1) ~) Attempt any Five questions. All questions carry equal marks.

Maximum Marks: 100

Ql) (a) Defme learning rule? Demonstrate-using an example to conclude the


.

comparison between the supervised andunsupe1irised learning algorithm?

(b) What is the significance of transfer function? List the different types of transfer functions? If the inputto a single input neuron is 2.0, its weight' is 1.3 and .its bias is 3.0.,what is t~e 'output of the neuron if it has the
followingtrans~er ~nctio1].?' (i) Hatdlimif (ii) Linear
.. ~ .. .

Q2) (a) Explain why a single-layer perceptron cannot solve the XOR problem. Use an Xl vs. ,Xl plot to show that a straight line cannot separate the XNOR states. List the several aspects to keep in mind when selecting an . appropriate' neural network structure. (b) Explain the major difference between a conventional (serial) computer and a neural network.

Q3) (a) Give the comparison between the radial basis-functionnetwor~s and the multilayer perceptron? Train the home made robot using recurrent back
propagati9n algorithm. ~

(b)

Explain in detail the procedure for designing the neural network using competitive learning? "

J-4043[8129]

P. T.o.

Q4) A liquid level system is illustratedin the followingfigur;. Th; obje;ti~eit =-system is to control the inJ>utflow such that desired heights of the fluid..in both tanks can be maihtained. ,A
~

Input Flo'vV

qi~J

.'

I~ w'

Al

'hI

Rl

~ql

~ J..
-.-- 2'

R2
~

112.

~OillpmFlow

,I 1

In the figure, qi: cil, q2 = the rates ofliquid flow,

,'I

; hI' h2 = the heights of fluid level, RI' Rz = the flow resistance, and AI, ~= the cross-sectiqp.al tank areas. , The following basic linear relationships are valid for this system,'namely,
q , h/r = rate of flow through orifice.

qn = (tank input rate of flow) - (tank output rate of flow)


"

==

net tank rate of flow.

(a) Develop.a neural m~tworkcontrol system that can effectively maintain theliqujd level heights in tanks 1 and 2. (b) Discuss th~ ~erf~~ance' of the ne~~~i~et\Vork'contr-~llerand :issues concerning fine tuning. Q5) In a water rig the control objective is'to adjust the water level in a tank after a step in the reference. The rig consists of a feed pump and a tank, and the tank has an outlet. (a) Assume first that the outlet is closed. As soon as the water reaches the setpoint level, the error becomes zero, and the. controller will stop the feed pump. (b) If there is overshoot, for example, if the feed pump reacts sluggishly,the controller should slow down the pump well before the water reaches the
~~cirt

I
1

(c) Ass,umenow that-the outlet IS open. .The controller must try and reach the setpoint and keep pumping to compensate for the water running out:
,

of the .outlet. A sustained control signal in steady state is necessary tc balance the outflow. Design a fuzzy logic_centrol for the above three problems using any (' the fuzzy P, PI, PD, Pill, PD+1 controller.

"

J-4043

- - -

,
~

Q6) (a) Discuss the training procedure of Kohenan policy? How to build prior
infor111a.fionn neural netWorks? i
,3!"

~b) Consider the logical Booleanfunction,A artdnot B, symbolically AnB', with fruth function

I xl=landx2 = 0 8 (xl' Xz) { 0 '" dlhe rwise


..

Design with.an appropriate neural networkto implement g.


Q 7) (a) (b) Dra\V'the block diagram of fuzzylogic .Explail1inbfiefthe basic concepts

of fuzzy logic controL it is closer to the real world..dive' reasons.


1~

Fuz;zylogic provides an alternative solution to non-linear control because


. .~.

Q8) ~rite'short notes on any three (a) Advantages offuzzy logic control over the artificial neural networks (b) Li1TI.itation oft}leIe9urrent backpropagafion a.lgorithm (c) Adaptive Resonance,Theory.
0'. (d) IopfieIcINetWorks;'

***

J-4043